Resilience scorecard

Sixteen questions about the parts of a home server that decide whether a bad day is an afternoon of work or a permanent loss. Each one is weighted by consequence rather than by effort, the weights add up to exactly 100, and the output is a score, a band, a breakdown by area and your open gaps ranked by what each costs you if it stays broken.

It is for the person whose setup works. Nothing here asks whether your services run, because they do. It asks the questions you can only answer honestly in private: whether the backup you configured in 2024 has been checked since, whether the port you forwarded to test something is still forwarded, and whether anybody would find out that a disk is dying before the array does.

Answer as things are today, not as you intend them to be. The three options are yes, partly and no, and partly is the correct answer far more often than people are comfortable with. Nothing you answer leaves the browser: there is no submission, no storage and no server side to this page.

How the score works#

The sixteen checks sit in five areas: backup (five questions, 42 points), exposure (four, 26), updates (three, 15), visibility (two, 11) and continuity (two, 6). Yes scores the full weight, partly scores half, no scores nothing. Because the weights total 100, your score is literally the sum of what you passed.

The weighting is the argument the page is making. "Every service you would miss is inside an automated backup" and "you have restored something in the last six months" are both worth 10, more than any security item, because they are the only two that decide whether data is recoverable at all. Off-site and non-deletable is worth 9. A UPS is worth 2: it prevents the most common cause of filesystem damage, and losing that fight costs you a fsck rather than your photos.

The gap list is sorted by points lost, not by difficulty, so the first item is frequently the annoying one. That is intentional. A list sorted by ease produces a stack of small wins and an untested backup.

What it deliberately does not model#

• Anything it can measure. It cannot see your machine, so every answer is self-reported. The scorecard is a structured way of being honest, not an audit.
• What your data is worth. Twenty years of family photos and a Jellyfin metadata database score identically. Weight the results yourself against what is actually irreplaceable.
• Quality within a yes. A backup that runs nightly to two destinations and one that runs weekly to a USB disk both score the same. The questions are deliberately coarse so that they are answerable.
• Business obligations. No retention regulation, no access logging, no key management beyond keeping secrets out of git. If you are running this for a business, treat a high score here as the floor.
• Physical and account security. Whether the server is in a locked room, and whether your registrar and email account have the recovery paths that would let somebody take your domain, are outside these sixteen questions and worth thinking about anyway.

What to do with the answer#

Fix one thing, today, and it should be the top of the ranked list. If that item is large, do the smallest honest version of it now and the full version later: restore a single file rather than rehearsing a whole recovery, back up the irreplaceable 200 GB before the 4 TB library, close the forwarded port before you design the perfect access story.

Then come back and re-run it. The Backup planner writes the job that closes most of the backup gaps, Backups that actually restore covers the rehearsal that closes the highest-weighted one, Remote access without port forwarding covers the exposure section, and The minimum viable monitoring stack covers finding out before your family does.